From: Arnd Bergmann <[email protected]>
Clang notices that there is a code path through
scmi_powercap_notify_supported() that returns an
undefined value:
drivers/firmware/arm_scmi/powercap.c:821:11: error: variable 'supported' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
821 | else if (evt_id == SCMI_EVENT_POWERCAP_MEASUREMENTS_CHANGED)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/firmware/arm_scmi/powercap.c:824:9: note: uninitialized use occurs here
824 | return supported;
| ^~~~~~~~~
drivers/firmware/arm_scmi/powercap.c:821:7: note: remove the 'if' if its condition is always true
821 | else if (evt_id == SCMI_EVENT_POWERCAP_MEASUREMENTS_CHANGED)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
822 | supported = dom_info->notify_powercap_measurement_change;
drivers/firmware/arm_scmi/powercap.c:811:16: note: initialize the variable 'supported' to silence this warning
811 | bool supported;
| ^
Return 'false' here, which is probably what was intended.
Fixes: c92a75fe84ce ("firmware: arm_scmi: Implement Powercap .is_notify_supported callback")
Signed-off-by: Arnd Bergmann <[email protected]>
---
drivers/firmware/arm_scmi/powercap.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/firmware/arm_scmi/powercap.c b/drivers/firmware/arm_scmi/powercap.c
index aae91f47303e..8ee3be8776b0 100644
--- a/drivers/firmware/arm_scmi/powercap.c
+++ b/drivers/firmware/arm_scmi/powercap.c
@@ -820,6 +820,8 @@ scmi_powercap_notify_supported(const struct scmi_protocol_handle *ph,
supported = dom_info->notify_powercap_cap_change;
else if (evt_id == SCMI_EVENT_POWERCAP_MEASUREMENTS_CHANGED)
supported = dom_info->notify_powercap_measurement_change;
+ else
+ supported = false;
return supported;
}
--
2.39.2
On Fri, Feb 16, 2024 at 05:32:53PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <[email protected]>
>
> Clang notices that there is a code path through
> scmi_powercap_notify_supported() that returns an
> undefined value:
>
Hi Arnd,
> drivers/firmware/arm_scmi/powercap.c:821:11: error: variable 'supported' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
> 821 | else if (evt_id == SCMI_EVENT_POWERCAP_MEASUREMENTS_CHANGED)
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> drivers/firmware/arm_scmi/powercap.c:824:9: note: uninitialized use occurs here
> 824 | return supported;
> | ^~~~~~~~~
> drivers/firmware/arm_scmi/powercap.c:821:7: note: remove the 'if' if its condition is always true
> 821 | else if (evt_id == SCMI_EVENT_POWERCAP_MEASUREMENTS_CHANGED)
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 822 | supported = dom_info->notify_powercap_measurement_change;
> drivers/firmware/arm_scmi/powercap.c:811:16: note: initialize the variable 'supported' to silence this warning
> 811 | bool supported;
> | ^
>
> Return 'false' here, which is probably what was intended.
>
> Fixes: c92a75fe84ce ("firmware: arm_scmi: Implement Powercap .is_notify_supported callback")
> Signed-off-by: Arnd Bergmann <[email protected]>
thanks for looking at this, this series that I've just posted is still
to be reviewd at all, so I would expect issues :D...BUT in this case I
dont think that the clang report is valid since, inside the culprit
function scmi_powercap_notify_supported(), a few lines before the
reported usage of unitialized data there is a check (@line 816) on the
'bounds' of evt_id itself
if (evt_id >= ARRAY_SIZE(evt_2_cmd) || src_id >= pi->num_domains)
return false;
so basically the mentioned if/else WILL be evaluated in some of its
branches for sure and supported wont be uninitialized.
Indeed, I removed from here (and from all the series) the explicit
initialization at definition time right before posting the series.
Having saidm that...maybe it is just brain-dead this approach of mine
since it is able to fool clang & friends...I would add bACK an explicit
initialization of supported all across this series in V2, if this
sounds good to you.
Thanks,
Cristian
> ---
> drivers/firmware/arm_scmi/powercap.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/firmware/arm_scmi/powercap.c b/drivers/firmware/arm_scmi/powercap.c
> index aae91f47303e..8ee3be8776b0 100644
> --- a/drivers/firmware/arm_scmi/powercap.c
> +++ b/drivers/firmware/arm_scmi/powercap.c
> @@ -820,6 +820,8 @@ scmi_powercap_notify_supported(const struct scmi_protocol_handle *ph,
> supported = dom_info->notify_powercap_cap_change;
> else if (evt_id == SCMI_EVENT_POWERCAP_MEASUREMENTS_CHANGED)
> supported = dom_info->notify_powercap_measurement_change;
> + else
> + supported = false;
>
> return supported;
> }
> --
> 2.39.2
>
On Fri, Feb 16, 2024, at 18:21, Cristian Marussi wrote:
> On Fri, Feb 16, 2024 at 05:32:53PM +0100, Arnd Bergmann wrote:
>> From: Arnd Bergmann <[email protected]>
>>
>> Clang notices that there is a code path through
>> scmi_powercap_notify_supported() that returns an
>> undefined value:
>>
>
> thanks for looking at this, this series that I've just posted is still
> to be reviewd at all, so I would expect issues :D...BUT in this case I
> dont think that the clang report is valid since, inside the culprit
> function scmi_powercap_notify_supported(), a few lines before the
> reported usage of unitialized data there is a check (@line 816) on the
> 'bounds' of evt_id itself
>
> if (evt_id >= ARRAY_SIZE(evt_2_cmd) || src_id >= pi->num_domains)
> return false;
>
> so basically the mentioned if/else WILL be evaluated in some of its
> branches for sure and supported wont be uninitialized.
>
> Indeed, I removed from here (and from all the series) the explicit
> initialization at definition time right before posting the series.
>
> Having saidm that...maybe it is just brain-dead this approach of mine
> since it is able to fool clang & friends...I would add bACK an explicit
> initialization of supported all across this series in V2, if this
> sounds good to you.
I'm fine with any solution that avoids the warning. I usually
prefer the explicit assignment where it's needed over having
it as part of the declaration, and in this case I would
probably pick a switch/case of a set of if/else fi/else
blocks
Arnd